feat: add change proposal for openspec init command - #2
Merged
Conversation
TabishB
changed the title
This was referenced Feb 18, 2026
Closed
SWolfeCAI
referenced
this pull request
in AnchorSpec/AnchorSpec
Mar 27, 2026
Remove telemetry and rebrand to AnchorSpec
StevenWolfe
referenced
this pull request
in AnchorSpec/AnchorSpec
Mar 27, 2026
Remove telemetry and rebrand to AnchorSpec
DumoeDss
referenced
this pull request
in DumoeDss/rasen
Jul 9, 2026
Moves the completed prompt-audit-fixes-verify-ship change (child #2 of the prompt-audit-fixes portfolio) to rasen/changes/archive/2026-07-09-prompt-audit-fixes-verify-ship/ and syncs its two delta specs as new capabilities in rasen/specs/: - verify-ship-evidence: the verify -> ship evidence chain — verify-change persists verification-report.md; both verify variants emit a shared canonical verdict + machine-checkable status line and record fingerprinted test evidence when they run tests; ship consumes these as verification and test-skip evidence; chrome-use added to the parity golden master. - expert-source-reading-scope: the "Never read source code" absolutes in QA_METHODOLOGY and DESIGN_METHODOLOGY scoped to the exploration/testing (audit) phase, with explicit carve-outs for diff-aware triage and the standalone fix loop. Staged the two new spec subdirectories individually (not the bare rasen/specs/ parent) per the established precise-pathspec discipline; confirmed no other content was staged under rasen/specs/ at commit time (git status --porcelain -- rasen/specs/ showed only these two untracked dirs before staging). All 19/19 tasks complete; all 4 artifacts done; both synced specs validate clean (`rasen spec validate`). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
DumoeDss
referenced
this pull request
in DumoeDss/rasen
Jul 9, 2026
…evival, two-threshold rule, counter disambiguation Closes a family of ambiguities in the LEAD orchestration playbook (_orchestration.ts, auto.ts, goal-command.ts) around worker death, threshold counters, and gate semantics — cases where the same-sounding number or event was governed by two different rules depending on context, with no explicit disambiguation. - OR-1/SH-4 threshold families: relay caps, stall limits, and round caps disambiguated — each named explicitly and scoped to the lifecycle event it governs. - H.4/OR-6 worker-death taxonomy + infra revival: two-branch taxonomy — (a) genuine work-exhaustion (HANDOFF) vs (b) infrastructure death (process/connection failure). An infra-death SendMessage-triggered revival is NOT charged against the relay/ stall caps that govern (a). The (a)-vs-(b) boundary has a documented residual ambiguity, accepted in design Risks as cheap/self-correcting (a genuinely-full worker HANDOFFs on the wake turn and gets reclassified). - OR-9/OR-13/OR-15 counter disambiguation: a counter table maps every counter token (maxRelays, stallLimit, maxRounds, loop.maxRounds, strategyAttempts, loopStallLimit, sessionHandoff.n) to its exact governing rule — grep-audited, no orphans. - Tier-C degradations (OR-4, OR-14): LEAD never self-certifies a Tier-C implementer's work without an independent check. verifyPolicy standard/light modes defined in auto.ts section 5 (verify stages only run under the auto command). - OR-8/OR-11/SH-5/SH-7: SH-5 scopes the LEAD's "flag it" duty — an out-of-scope issue the LEAD notices routes to DONE durable-findings, not autonomous investigate/fix/ask-user; explicitly disclaims reopening child #1's dispatched-mode report-only contract. SH-7 defines a narrow inline-fix exception. - Gate directive precedence (D6, LEAD ruling): Step G resolves how a decomposed portfolio's child-pipeline gates behave with no explicit parent gate directive — default is auto-continue (record the checkpoint, do not pause per child), since irreversible actions already collapse to the single portfolio-level delivery decision. Reversible product decision, surfaced to the user in the ship log rather than silently baked in. Round 2 tightened Step G's prose after the reviewer flagged that the wording could be misread as the opposite default — behavior unchanged, wording clarified. _orchestration.ts/auto.ts/goal-command.ts are NOT in the parity golden-master registry (pre-existing gap). Verified instead via node build.js + node dist/cli/index.js update succeeding, plus 5 independent grep spot-checks against the regenerated rasen-auto/ rasen-review-cycle/commands-auto output confirming the new prose is present verbatim, performed independently by implementer and reviewer. Registry expansion relayed to child #5 (prompt-audit-fixes-lifecycle), along with a newly-discovered, distinct gap: rasen-goal/goal-plan templates are absent from the generation registry entirely, not just the parity registry. Installed skills regenerated via node build.js. Parity hashes resynced for the 15 PREAMBLE-importing expert templates (careful/freeze/guard/unfreeze correctly show zero movement — they do not import PREAMBLE; a task-note miscount of "19" was corrected to 15 in fix round 1). Reviewed CLEAN APPROVED (review-cycle, 2 rounds, non-author confirmed). Round 1: 1 Minor fixed (auto.ts's gate guardrail lacked a back-reference to Step G's precedence rule — added a scoped cross-reference), 1 Trivial fixed (tasks.md miscounted PREAMBLE embedders as 19, corrected to 15). Round 2: 1 new Minor found and fixed (Step G sentence 1's "does NOT suppress ... by default" wording could be misread as contradicting the D6 auto-continue default — tightened without changing behavior). Seam integrity confirmed: child #1's Step B report-contract sentence and child #2's verify/ship evidence chain both verified untouched by this diff. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
DumoeDss
referenced
this pull request
in DumoeDss/rasen
Jul 9, 2026
…er, archive hard gates, delivery preconditions, F.1 generation-match Wires the propose/office-hours/archive/apply/continue workflow commands into the evidence chain and gates established by children #1-#3, and closes a resume-ladder generation-match gap in orchestration. - WF-2: propose.ts checks both office-hours-design.md (change dir) and the sibling office-hours directory (by slug) before drafting, incorporating found validation and naming office-hours as the source; all paths resolved from status JSON. office-hours.ts's producer-side notes (dual-write, auto-detect, downstream integration) made accurate to match, with a verbatim slug-derivation example matching propose's so the two-sided derivation converges (fixed in round 1 after the reviewer caught divergent examples). - WF-6: office-hours.ts (workflow command) restructured so the /office-hours expert is the single facilitation authority; the inline six-questions/builder text becomes an explicit fallback pre-brief, not a second pass; doc production consolidated to one step; precedence stated. - WF-4: archive-change.ts gates on verification-report.md's VERIFY VERDICT (child #2's evidence file) -- refuses by default on BLOCKED, requires an explicit blocker-naming override, refuses outright non-interactively; CLEAN passes with no gate; absence gets a soft note only. Incomplete-tasks check elevated from soft warning to the same hard-gate semantics. The "don't block on warnings" guardrail re-scoped (enumerate-and-gate) to soft warnings only, explicitly excluding both new hard gates. - WF-5/WF-11: archive-change.ts adds a delivery-precondition check reading ship-log.md (soft warning if absent, with a spec-only escape) and a portfolio-awareness note when the ship log's Status line reads "delivery deferred to portfolio level" (ship.ts's exact local-mode marker). - WF-5 (apply) / WF-10 (continue): completion nudges in apply-change.ts and continue-change.ts steer through /rasen:verify -> /rasen:ship, naming archive only as the post-delivery step rather than an immediate co-equal option. - F.1 (LEAD-added, _orchestration.ts scope exception -- one clause, nothing else touched, children #1/#3's edits survive byte-for-byte): Step F.1's resume ladder now states a handoff document counts only if it's the LATEST holder's own distillation; an un-exhausted latest holder with no document resumes from its TRANSCRIPT instead, which beats any earlier generation's document. Adds the same-session- restart nuance: SendMessage-by-NAME may still resolve the latest holder post-restart -- try that wake first, fall back to the ladder otherwise. This exact behavior was live-validated during this change's own implementation (see ship-log's Infra-Revival section). - Parity registry expansion: +19 function-hash entries + 11 content-hash entries for the previously-unpinned workflow/command family (office-hours-command, verify-enhanced, ship, retro, auto, review-cycle, handoff, goal-plan/iterate/report/command) -- this is what makes the F.1 edit verifiable, since it hash-locks rasen-auto/rasen-goal/rasen-review-cycle for the first time. Reviewed CLEAN (review-cycle round 1, non-author confirmed): 1 Minor fixed (propose/office-hours slug-example divergence), 1 Minor accepted-known (F.1's hash movement can't be diff-verified against a pre-edit baseline since the templates were pinned in the same regen run as the edit -- D8, inherent to prior parity debt, not a defect), 2 Trivial accepted-known and skipped by design (office-hours wording nuance deferred to child #6's WF-3; F.1 clause density -- tightening declined to avoid re-churning newly-locked hashes for zero functional gain). Seam integrity confirmed byte-compatible both directions against the actual producer strings, called out by the reviewer as "the opposite of the WF-1 disease" this portfolio exists to close. Infra-revival event during fix round 1: implementer's connection closed mid-edit; LEAD applied child #3's H.4(b) protocol (SendMessage-revive the same agent, zero rework since edits had already landed) rather than cold-reconstructing -- not charged to relay/stall budget, and incidentally a live validation of this change's own F.1 same-session-restart nuance before its hash-lock even shipped. Known-open item relayed to the portfolio-end report (not this change's scope): goal templates are registered in the generation registry, but `update` still doesn't emit rasen-goal* skill directories -- suspected deploy-path filter, not root-caused. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
DumoeDss
referenced
this pull request
in DumoeDss/rasen
Jul 9, 2026
Moves the completed prompt-audit-fixes-lifecycle change (child #5 of the prompt-audit-fixes portfolio) to rasen/changes/archive/2026-07-09-prompt-audit-fixes-lifecycle/ and syncs its six delta specs into rasen/specs/: NEW capabilities: - lifecycle-stage-sequencing: apply/continue completion messages steer through the intended stage sequence (verify -> ship -> archive) instead of offering archive as an immediate co-equal step. - workflow-template-parity: the workflow/orchestration skill and command templates outside the expert set are now pinned by the parity golden master, mirroring the expert templates and chrome-use. MODIFIED capabilities: - propose-workflow (ADDED-only): consumes office-hours validation output as input context, checking both the change dir and the sibling office-hours directory by slug. - opsx-office-hours-command (ADDED-only): facilitation delegates to the /office-hours expert as the single authority; inline six-questions/builder text becomes a fallback pre-brief only. - orchestration-worker-lifecycle (ADDED-only): F.1 resume ladder prefers the latest holder's own distillation over any earlier generation's document; an un-exhausted latest holder with no document resumes from its transcript instead. - opsx-archive-skill (deliberate MODIFY, not ADDED-only): the Task Completion Check requirement is REPLACED wholesale -- incomplete tasks upgraded from a soft warning to a hard gate (refuse by default, explicit override required, refuse outright non-interactively). Two new ADDED requirements inserted alongside it: Verification Verdict Gate (hard-gates on VERIFY VERDICT: BLOCKED, consuming child #2's evidence file) and Delivery Precondition Check (soft warning on missing/portfolio-deferred ship-log.md). All six specs validated clean (`rasen spec validate`). Staged the six spec subdirectories individually (not the bare rasen/specs/ parent) per the established precise-pathspec discipline; confirmed git status --porcelain -- rasen/specs/ showed exactly these six entries (4 modified, 2 untracked new) before staging, nothing from another session. All 20/20 tasks complete; all 4 artifacts done. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Clarify that patches are required for ALL spec changes, including brand new capabilities